Run 6. Outer Step 43. Inner Step 4.

config.json

@@ -76,7 +76,7 @@
     "163": "NON_PARTICIPATING",
     "164": "NOT_ALIVE",
     "165": "NON_PARTICIPATING",
-    "166": "NOT_ALIVE",
+    "166": "SUCCESS",
     "167": "NOT_ALIVE",
     "168": "NOT_ALIVE",
     "169": "NOT_ALIVE",
@@ -104,7 +104,7 @@
     "189": "NOT_ALIVE",
     "19": "NOT_ALIVE",
     "190": "SUCCESS",
-    "191": "NOT_ALIVE",
+    "191": "NON_PARTICIPATING",
     "192": "NOT_ALIVE",
     "193": "SUCCESS",
     "194": "NOT_ALIVE",
@@ -113,7 +113,7 @@
     "197": "NOT_ALIVE",
     "198": "NOT_ALIVE",
     "199": "SUCCESS",
-    "2": "SUCCESS",
+    "2": "NOT_ALIVE",
     "20": "NOT_ALIVE",
     "200": "NOT_ALIVE",
     "201": "SUCCESS",
@@ -177,22 +177,22 @@
     "254": "NOT_ALIVE",
     "255": "NOT_ALIVE",
     "26": "NON_PARTICIPATING",
-    "27": "NON_PARTICIPATING",
+    "27": "NOT_ALIVE",
     "28": "NON_PARTICIPATING",
     "29": "NOT_ALIVE",
     "3": "SUCCESS",
     "30": "NOT_ALIVE",
-    "31": "SUCCESS",
+    "31": "NOT_ALIVE",
     "32": "SUCCESS",
     "33": "NOT_ALIVE",
     "34": "NOT_ALIVE",
-    "35": "SUCCESS",
+    "35": "NOT_ALIVE",
     "36": "NOT_ALIVE",
     "37": "NOT_ALIVE",
     "38": "NOT_ALIVE",
     "39": "NOT_ALIVE",
     "4": "NOT_ALIVE",
-    "40": "NOT_ALIVE",
+    "40": "SUCCESS",
     "41": "NOT_ALIVE",
     "42": "NOT_ALIVE",
     "43": "NOT_ALIVE",
@@ -209,24 +209,24 @@
     "53": "SUCCESS",
     "54": "NOT_ALIVE",
     "55": "NOT_ALIVE",
-    "56": "NON_PARTICIPATING",
+    "56": "NOT_ALIVE",
     "57": "NON_PARTICIPATING",
     "58": "NOT_ALIVE",
     "59": "NOT_ALIVE",
     "6": "NOT_ALIVE",
-    "60": "NON_PARTICIPATING",
+    "60": "NOT_ALIVE",
     "61": "NOT_ALIVE",
     "62": "NOT_ALIVE",
     "63": "NOT_ALIVE",
     "64": "NOT_ALIVE",
-    "65": "NOT_ALIVE",
+    "65": "NON_PARTICIPATING",
     "66": "NOT_ALIVE",
     "67": "NOT_ALIVE",
     "68": "NOT_ALIVE",
     "69": "NOT_ALIVE",
     "7": "NOT_ALIVE",
     "70": "SUCCESS",
-    "71": "NON_PARTICIPATING",
+    "71": "FAIL",
     "72": "NOT_ALIVE",
     "73": "NOT_ALIVE",
     "74": "NOT_ALIVE",
@@ -253,7 +253,7 @@
     "93": "NOT_ALIVE",
     "94": "NOT_ALIVE",
     "95": "NON_PARTICIPATING",
-    "96": "NOT_ALIVE",
+    "96": "NON_PARTICIPATING",
     "97": "NOT_ALIVE",
     "98": "NOT_ALIVE",
     "99": "SUCCESS"
@@ -264,17 +264,17 @@
   "attention_bias": false,
   "attention_dropout": 0.0,
   "block_list": [
-    6157525,
-    6157542
+    6157626,
+    6157644
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 2048,
   "initializer_range": 0.02,
-  "inner_step": 76,
+  "inner_step": 4,
   "intermediate_size": 5632,
-  "last_allreduce_block": 6156907,
+  "last_allreduce_block": 6157557,
   "max_position_embeddings": 2048,
   "mlp_bias": false,
   "model_type": "llama",

gradients.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4963cbccff8c5493eec4413f16d5be56f17aa2af43aafff20682e9b9e981268c
+oid sha256:332e2900d104ede530ddff3993f90cb1d0428d43da489607fa3d3c7702177d58
 size 88389626

inner_optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3f15727d922e2bfb5ec2f46e8c88a2fead97c42f472c95c9ad11c50f8396982
+oid sha256:cc4125ee66076d1e9b6200433724dfb95d4c4fa2a1dcbc0459a5d81731e9997b
 size 8800564076

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21d3f84ea3e8910f022d780731633a4b7a11caa5b44d7e7f20c82ce313e92cca
+oid sha256:1c6c4c5b50ed603c09ad1c70f7150eaf17068beec447cef9592cdf553974376e
 size 4400216536